1 Introduction

Hydrological changes associated with earthquakes have been observed at many places in the world. Generally, changes in groundwater level occur co-seismically (Waller 1966; Roeloffs 1998; King and Igarashi 1999; Chia et al. 2001; Cox et al. 2012; Weingarten and Ge 2014, He et al. 2017), while changes in streamflow or spring flow appear after major earthquakes (Rojstaczer and Wolf 1992; Muir-Wood and King 1993; King et al. 1994; Manga 2001; Manga et al. 2003; Montgomery et al. 2003; Wang et al. 2004a, b; Manga and Rowland 2009; Mohr et al. 2012, 2017).

Earthquake-triggered hydrological changes may provide insight into the mechanism underlying the crustal processes during the fault rupture. Groundwater-level changes in the vicinity of seismogenic fault are generally attributed to the redistribution of stress–strain due to fault movement (Roeloffs 1988; Montgomery and Manga 2003). Rojstaczer and Wolf (1992) proposed that streamflow increases are likely caused by the increase in permeability of near-surface rock due to fracturing by seismic shaking. Muir-Wood and King (1993) related the hydrological changes following major earthquakes to the style of faulting. Significant streamflow increases were found to accompany major normal fault earthquakes, while small decreases were associated with reverse fault earthquakes. Soil liquefaction has been proposed for the streamflow increase (Manga 2001; Wang et al. 2001; Montgomery et al. 2003), but possibly limited to shallow sediments in the river valley of the mountain area (Mohr et al. 2012).

During the 1999 M W 7.6 Chi–Chi earthquake, the largest and most destructive inland earthquake in the history of Taiwan, stream discharges recorded by a dense network of gauges on major rivers and their tributaries provide comprehensive data for exploring earthquake-triggered hydrological changes. In this paper, we presented the streamflow changes in the vicinity of the seismogenic fault before and after the Chi–Chi earthquake. Long-term data analysis and further investigations provide a basis for a better understanding of the natural processes of the earthquake hydrological anomalies.

2 Seismicity, Geology and Hydrology

As a part of the Circum-Pacific seismic belt, Taiwan is located at the convergent boundary between the Philippine Sea plate and the Eurasian plate. As a result of movement and collisions of the two plates, Taiwan is one of the most seismically active regions in the world. Historical destructive earthquakes could be documented back to 1644, while earthquakes recorded by seismic instruments started after 1897 (Hsu 1980). From 1990 to 2006, there were 899 earthquakes cataloged with magnitude greater than M W 5.5 (Chen and Tsai 2008).

The island of Taiwan is located at the margin of young orogenic belt containing a series of north-northeast trend geologic formations (Ho 1986). The western plain consists of a thick sequence of undeformed unconsolidated deposits, while sedimentary and metamorphic rocks appear along with complex fold and fault structures in the mountainous area. As mountain ranges in the island stretch from north to south, most rivers flow in the east–west direction. Groundwater storage is different in the consolidated formation and the unconsolidated sediments (WRA 2016). In the mountains area, groundwater is stored in the pores of sedimentary rocks as well as fractures, such as joints and faults, of the metamorphic rocks. Major aquifers, however, are located in gravel and sand layers of unconsolidated deposits in the western plain.

3 Chi–Chi Earthquake and Co-seismic Groundwater-Level Changes

The 1999 Chi–Chi earthquake occurred at 1:47 am on September 21, 1999, local time (17:47 on 20 September UTC). The epicenter of the M W 7.6 earthquake is located at 23.85°N; 120.82°E, near the town of Chi–Chi in central Taiwan (Fig. 1). The depth of the hypocenter of the earthquake is about 8 km. The focal mechanism of the mainshock was of a thrust type with strike 5°, dip 34°. Surface rupture appeared to extend approximately 100 km in the north–south direction along the traces of the Chelungpu thrust fault (Angeliera et al. 2003). Horizontal and vertical offset caused by crustal deformation ranged from 2.4 to 10.1 m and 1.2 to 4.4 m across the Chelungpu fault, respectively. In the epicentral region, the crustal deformation was essentially a uniaxial compressional strain of 0.36 micro-strain/year in the direction of 114° over several years before the earthquake (Yu et al. 2001). After the mainshock, more than 10,000 aftershocks were recorded in the first 3 weeks (Ma et al. 1999).

Fig. 1
figure 1

Distribution of hydrological monitoring stations during the 1999 M W 7.6 Chi–Chi earthquake in central Taiwan. The upper left inset shows the area in Taiwan. Post-earthquake streamflow increases (blue circles) and decrease (red circle) were observed at 23 stream gauges. Co-seismic groundwater-level rises (blue triangles) and falls (orange inverted triangles) were observed in 152 wells. The inset map in the lower right shows the location of gauge stations CS64, CS44, CS63 and the earthquake epicenter

Full size image

In the previous studies, examination of available data recorded at monitoring wells in the footwall of the Chelungpu fault revealed widespread co-seismic groundwater-level changes (Wang et al. 2001). Widespread sustained co-seismic groundwater-level changes were observed in monitoring wells in the footwall (Chia et al. 2001, 2008a; Wang et al. 2001). As shown in Fig. 1, groundwater-level falls were recorded in wells near the ruptured fault, while rises were recorded in most wells farther away to the west in the coastal plain consisting of unconsolidated deposits (Chia et al. 2001). Generally, larger changes, up to 11.09 m, were found in wells closer to the fault. These co-seismic changes were likely caused by co-seismic strain induced by fault displacement (Wakita 1975; Roeloffs 1988; Grecksch et al. 1999; Ge and Stover 2000; Chia et al. 2001, 2008b). The distribution of the sustained changes in the footwall revealed that crustal extension dominated near the Chelungpu fault during the earthquake, while compression prevailed away from the fault.

4 Rainfall and Stream Gauges

In central Taiwan, the average annual precipitation is about 2154 mm, but it may reach to 4000 mm in the mountainous area. The climate of the area is characterized by a distinct wet and dry season, with most of the rainfall occurring between May and September. Heavy rainfall usually occurs in short duration, and thus, most stream discharges peak shortly after rainfall. The river usually takes a few days to return to the base flow. Significant rainfall-induced changes in streamflow are frequently recorded during wet seasons. The Chi–Chi earthquake occurred at the end of the wet season. During the dry season, the gradual decay in discharge reflects diminishing supply from the groundwater discharge.

During the Chi–Chi earthquake, streamflow discharges were recorded at 32 gauges on four major rivers, including Daan River, Dajia River, Wu River and Choshui River, and their tributaries in the vicinity of the Chelungpu fault (Fig. 1). The four rivers originated from the Central Mountain Ranges, flowing westward through the coastal plain to the Taiwan Strait. The length of these rivers ranges from 96 to 187 km, and the watershed area ranges from 758 to 3157 km2 (Table 1). Of those, the Choshui River is the longest river in Taiwan.

Table 1 Stream gauges on four rivers in central Taiwan
Full size table

These gauges were installed by the Water Resource Agency for the water resource management and flood prevention. Most of these gauges are located on the mountainous east, or hanging wall, side of the fault. They were used to measure the elevation of the stream water surface. Water level sensor system was designed to fit the characteristics of the cross section of the river at these gauges. A water-stage recorder was installed on or near the bridge crossing the river for automatically recording water level of the stream. The water level was then converted to the discharge (volumetric flow rate) based on the stage-discharge relation. Currently, the river water level at these gauges is recorded at 1-h intervals. In 1999, however, the water level was recorded daily, with the exception of the CS44 gauge where the data were recorded at 1-h intervals.

5 Streamflow Changes Induced by the Earthquake

Streamflow changes recorded at some of these gauges after the Chi–Chi earthquake were reported before (Wang et al. 2004b). As additional data became available and further investigations were conducted, it was found that, among the streamflow changes recorded at 32 changes, six anomalously large increases were caused by upstream reservoir release for the dam safety, one substantial decrease was caused by a large landslide dam on the river due to the earthquake, and two changes were indistinguishable from rainfall disturbances or daily fluctuation. Thus, only 23 gauges recorded distinct earthquake-related changes. Of those, 22 are post-earthquake streamflow increases and the remaining one is post-earthquake decrease (Table 2). The distance between these gauges and the epicenter ranges from 4 to 69 km.

Table 2 Post-earthquake streamflow increase at 22 stream gauges
Full size table

5.1 Post-earthquake Streamflow Increase

Six gauges on the Dajia River and its tributaries recorded a streamflow increase, ranging from 58 to 833%, in 2 to 3 days after the earthquake (Table 2). The rapid increase was followed by a slow decay over the next month or more. The temporal variations of daily discharge at these gauges are shown in hydrographs, along with nearby rainfall data (Fig. 2). The largest percentage increase was recorded at DJ46, with an elevation of 550 m, located on a tributary in the downstream of the Dajia River. The discharge increased 833% to 2.8 m3/s 3 days after the earthquake. The large percentage increase was primarily caused by the relatively small discharge, approximately 0.3 m3/s, on the day before the earthquake. The other five gauges are located near the source of the Dajia River with an elevation between 1434 and 1629 m. For instance, DJ35 is located on the mainstream for monitoring the sum of upstream tributary discharges. An increase of 83% in discharge, from 16.5 m3/s on the day before the earthquake to a peak flow of 30.2 m3/s 3 days after the earthquake, was recorded at the gauge (Fig. 2). The peak discharge took more than 1 month to return to the pre-earthquake value. The slow decay could be disturbed by rainfall, such as an increase in discharge following the heavy rainfall in mid-December.

Fig. 2
figure 2

Daily discharge from August to December 1999 at six gauges on the Dajia River and its tributaries. Streamflow increases, ranging from 58 to 833%, were recorded after the earthquake

Full size image

The temporal variations of daily discharge at five gauges on the Wu River and its tributaries show streamflow increases, ranging from 63 to 413%, after the earthquake (Fig. 3). At WU28, an increase of 413% in discharge was recorded on the mainstream. The discharge increased from 9.6 m3/s on the day before the earthquake to a peak value of 49.2 m3/s 2 days after the earthquake. The WU25 gauge, with an elevation of 10 m, is located on the mainstream in the coastal plain. It recorded a streamflow increase of 63%, from 103.0 m3/s to 168.0 m3/s, in 3 days after the earthquake.

Fig. 3
figure 3

Daily discharge from August to December 1999 at five gauges on the Wu River and its tributaries. Streamflow increases, ranging from 63 to 413%, were recorded after the earthquake

Full size image

Four gauges on the Daan River and its tributaries recorded a streamflow increase, ranging from 139 to 311%, 1 to 4 days without rainfall after the earthquake (Table 2). The temporal variations of daily discharge at these gauges are shown in Fig. 4. At gauge DA12, for instance, an increase of 214% in discharge was recorded on a small tributary. The discharge increased from 2.1 m3/s on the day before the earthquake to a peak value of 6.6 m3/s 4 days afterward. The peak flow took about 2 months to return to the pre-earthquake level. At DA15, the discharge increased to the first peak value of 34.1 m3/s 2 days after the earthquake and then decayed during the third day. Instead of a gradual decay, the discharge increased again for a week to the second peak. Further investigations indicated that several small landslide dams and lakes were created in the upstream of DA15 during the earthquake (Lin et al. 2000; Chen and Shi 2000). The collapse of these dams and the rainfall at end of September provided the discharge for the second increase to the peak of 52.8 m3/s on October 1. A similar pattern of post-earthquake streamflow changes was observed at DA11, which is located in the downstream of DA15.

Fig. 4
figure 4

Daily discharge from August to December 1999 at four gauges on the Daan River and its tributaries. Streamflow increases, ranging from 139 to 311%, were recorded after the earthquake

Full size image

Seven gauges on Choshui River and its tributaries recorded streamflow increases after the earthquake (Fig. 5). Of those, four gauges, CS11, CS63, CS57 and CS58, were installed on the mainstream, while the other three, CS40, CS48 and CS49, were on the tributaries where the discharge is relatively smaller. The streamflow increases at these gauges, ranging from 61 to 143%, are listed in Table 2. For instance, at CS48, the discharge increased from 26.4 m3/s on the day before the earthquake to a peak value of 42.4 m3/s, or an increase of 61%, over 2 days after the earthquake. The increased streamflow took about one and a half month to return to the pre-earthquake level. At CS11 and CS49, however, two post-earthquake peak flows were recorded after the earthquake. Further investigations indicated the anomalous changes were also caused by the collapse of landslide dams and local rainfall in the upstream.

Fig. 5
figure 5

Daily discharge from September to December 1999 at seven gauges on the Choshui River and its tributaries. Streamflow increases, ranging from 61 to 143%, were recorded after the earthquake

Full size image

5.2 Post-earthquake Streamflow Decrease

After the 1999 Chi–Chi earthquake, the only streamflow decrease was observed at gauge CS64. This gauge, located only 4 km to the southeast of the epicenter, was placed to monitor streamflow in the downstream of the Shueili Creek (inset of Fig. 1). The Shueili Creek, a tributary of the Choshui River, lies almost directly above the hypocentral area and coincides approximately with the Shueilikun fault. Its bedrock is composed of fractured quartzitic sandstone and arkosic sandstone interbedded with argillite. The discharge recorded at CS64, as shown in Fig. 6a, displayed a large and abrupt decrease, from 162.0 to 58.8 m3/s, on the day of the Chi–Chi earthquake. During the following few days, it further decreased to 7.1 m3/s.

Fig. 6
figure 6

Time series of daily discharges recorded between August and December 1999 at CS64 and CS44. a The daily hydrograph at CS64 (orange line) showing the discharge decreased 96% immediately after the earthquake. b The daily (blue line) and hourly (gray line) hydrograph at CS44 showing the discharge was primarily affected by occasional water releases from the nearby Mingtan reservoir on the Shueili Creek

Full size image

There was another gauge, CS44, located near the base of the Mingtan Dam approximately 3-km upstream from CS64 along the Shueili Creek (inset of Fig. 1). CS44 was installed by the Taiwan Power Company to monitor the release of Mingtan reservoir to the Shueili Creek. The hourly discharge at CS44, as shown in Fig. 6b, was normally maintained below 50 m3/s, except for emergency need or during heavy rainfall. It did not show either an abrupt post-earthquake decrease, as recorded by CS64. In fact, the daily discharge at CS44 increased from 14.1 m3/s immediately before the earthquake to 89.6 m3/s after the earthquake. The hourly discharge also showed a rapid increase, up to 202.5 m3/s, after the earthquake (Fig. 6b). The increase in discharge or reservoir release was imposed for dam safety. However, such a large increase at CS44 did not cause an increase in discharge at the downstream gauge CS64. Instead, the post-earthquake discharge continued to decline.

The long-term data provide a more detailed comparison in discharge between the gauges CS64 and CS44, as shown in Fig. 7a. Generally, the discharge at the two gauges shows a similar seasonal pattern that increases in the wet seasons and decreases in the dry seasons. It is also noted that, similar to most rivers that obtain their water from the groundwater discharge during the dry seasons, the discharge of the Shueili Creek at the downstream gauge CS64 was usually larger than that at the upstream gauge CS44. However, a reversed situation appeared between September 1999 and June 2000 (Fig. 7b). The reversal event suggests that the Shueili Creek was changed to a influent (losing) stream from a effluent (gaining) stream immediately after the occurrence of the earthquake on September 21, 1999. Such a phenomenon lasts until the wet season arrived in May 2000. Therefore, based on the long-term monitoring data, the Shueili Creek had lost water to the creek bed over the 3-km segment between the gauges CS44 and CS64 in the 8 months after the earthquake.

Fig. 7
figure 7

Daily discharges at two gauges, CS64 and CS44, on the Shueili Creek from 1994 to 2004. a Downstream discharges at CS64 (orange line) and upstream discharges at CS44 (blue line). b The difference in discharges between CS64 and CS44. The daily discharge at CS64 is normally larger than that at CS44 upstream (green color), but the situation was reversed after the Chi–Chi earthquake for 8 months (red color), indicating creek water loss to the ground between CS44 and CS64

Full size image

Near CS64 was the other gauge, CS63, which recorded no streamflow decrease, but a post-earthquake increase, like most other gauges (Fig. 5). This is not surprising, because CS63 is located on the mainstream of the Choshui River, the recorded data represent the sum of discharges from all upstream tributaries (inset of Fig. 1).

6 Discussions

The post-earthquake streamflow increases recorded by the 22 gauges are similar to those observed in the previous studies. The only source of the increases with no rainfall comes from the discharge of groundwater. Based on Darcy’s law, the increase of groundwater discharge to the river is caused by either an increase of rock permeability in the hills or a rise of pore pressure in the ground. The sudden increase of permeability is often explained by rock fracturing due to seismic shaking (Rojstaczer and Wolf 1992; Rojstaczer et al. 1995; Sato et al. 2000; Wang et al. 2004a, b; Charmoille et al. 2005; Elkhoury et al. 2006; Wang and Manga 2010; Manga et al. 2012; Mohr et al. 2015). Such an explanation can be supported by the occurrence of numerous landslides in the mountain area during the 1999 Chi–Chi earthquake (Hung 2000). For instance, the removal of vegetation and subsoil took place at 99 peaks during the earthquake, exposed the bedrock to the surface and enabled the groundwater to move through fractures more easily. Part of the streamflow increase may also be attributed to a rise of pore pressure induced by compressive strain during the fault rupture. Such an increase can be supported by the co-seismic groundwater-level rise prevailed in the footwall or the plain area. The mountain ranges in Taiwan were formed primarily by the compression between the Eurasian plate and the Philippine Sea plate. However, groundwater-level data were not recorded in the mountainous hanging wall during the Chi–Chi earthquake to support the compressive strain.

The earthquake-triggered streamflow increase is similar to the rainfall-induced increase, typically peaked in a few days. However, unlike a rapid decrease in discharge following the peak flow after rainfall, the post-earthquake peak flow is followed by a slow decay which may last for several months (Figs. 2, 3, 4). The slow decay is attributed to a prolonged groundwater discharge increase, instead of runoff, to the river. Such a groundwater discharge increase is consistent with an increase in either permeability or pore pressure proposed for the post-earthquake streamflow increase.

A large streamflow decrease triggered by the earthquake is rarely observed. Muir-Wood and King (1993) suggested that hydrological changes accompanying thrust fault earthquakes are most notable by their absence, or a decrease of well water level and spring flow. The field observation in the vicinity of the Chelungpu thrust fault, however, indicated post-earthquake streamflow increases at most gauges in the hanging wall. The only post-earthquake decrease in streamflow was recorded at the gauge nearest to the epicenter. Moreover, co-seismic falls in groundwater level were found primarily at wells adjacent to the thrust fault. Therefore, the crustal extension was likely to dominate from the epicentral region to the area adjacent to the thrust fault during the earthquake.

The long-term data indicate that the discharge at CS64 in the downstream of the Shueili Creek is usually larger than that at CS44 in the upstream. After the earthquake, however, the discharge at CS64 reduced to a level smaller than that at CS44 for 8 months or more. Such a post-seismic streamflow decrease might have been caused by a co-seismic decrease in pore pressure induced by crustal extension near the earthquake epicenter during the rupture of the thrust fault. The sudden decrease created a rapid downward discharge to the crust through the opening of pre-existing fractures in the bedrock of the Shueili Creek valley and new fractures generated by the earthquake. Such a downward discharge due to crustal extension must be far greater than the recharge from nearby hills due to seismic shaking. The sustained downward flow from surface water to the crust along the creek is likely to result in a streamflow in the downstream smaller than that in the upstream. The crustal extension is likely to extend to the area adjacent to the thrust fault, where co-seismic groundwater-level falls were recorded.

7 Conclusions

The significant streamflow decrease at CS64 on the Shueili Creek after the Chi–Chi earthquake is a unique observation. The gauge is only 4 km from the earthquake epicenter, and the Shueili Creek lies just above the hypocentral area. The proximity of gauge CS64 to the epicenter suggests that the streamflow decrease is likely induced by the crustal extension in the hanging wall during the rupture of the thrust fault. In the previous study (Chia et al. 2001), the co-seismic falls in groundwater level in the footwall near the Chelungpu fault trace were also attributed to the extensional deformation. Apparently, these hydrological anomalies reveal the dominance of the crustal extension in the area adjacent to the thrust fault in both the hanging wall and the footwall during the earthquake.

Beyond the zone of extension in central Taiwan, post-earthquake streamflow increase was recorded at all stream gauges and co-seismic groundwater-level rise was observed in nearly all monitoring wells. While the groundwater-level rise in the footwall was attributed to the compressive deformation, the streamflow increase in the hanging wall is possibly caused by both the permeability increase due to rock fracturing by seismic shaking and the pore pressure rise due to compressive deformation induced by fault displacement.